population pharmacokinetics‐pharmacodynamics of ... · population...

Population pharmacokinetics-pharmacodynamics ofvedolizumab in patients with ulcerative colitis and Crohn’sdiseaseM. Rosario*, N. L. Dirks†, M. R. Gastonguay†, A. A. Fasanmade*, T. Wyant*, A. Parikh‡, W. J. Sandborn§,B. G. Feagan¶, W. Reinisch**,†† & I. Fox*

*Takeda Pharmaceuticals InternationalCo., Cambridge, MA, USA.†Metrum Research Group LLC,Tariffville, CT, USA.‡Takeda Pharmaceuticals International,Inc., Deerfield, IL, USA.§Division of Gastroenterology,University of California San Diego, LaJolla, CA, USA.¶Robarts Research Institute, Universityof Western Ontario, London, ON,Canada.**Department Internal Medicine III,Medical University of Vienna, Vienna,Austria.††Department of Internal Medicine,McMaster University, Hamilton, ON,Canada.

Correspondence to:Dr M. Rosario, Clinical PharmacologyDepartment, Takeda PharmaceuticalsInternational Co., 35 LandsdowneStreet, Cambridge, MA 02139, USA.E-mail: [email protected]

Publication dataSubmitted 28 January 2015First decision 12 February 2015Resubmitted 24 April 2015Accepted 24 April 2015EV Pub Online 20 May 2015

This article was accepted for publicationafter full peer-review.

SUMMARY

BackgroundVedolizumab, an anti-a4b7 integrin monoclonal antibody (mAb), is indicated fortreating patients with moderately to severely active ulcerative colitis (UC) and Cro-hn’s disease (CD). As higher therapeutic mAb concentrations have been associatedwith greater efficacy in inflammatory bowel disease, understanding determinants ofvedolizumab clearance may help to optimise dosing.

AimsTo characterise vedolizumab pharmacokinetics in patients with UC and CD, toidentify clinically relevant determinants of vedolizumab clearance, and to describethe pharmacokinetic–pharmacodynamic relationship using population modelling.

MethodsData from a phase 1 healthy volunteer study, a phase 2 UC study, and 3 phase 3 UC/CDstudies were included. Population pharmacokinetic analysis for repeated measures wasconducted using nonlinear mixed effects modelling. Results from the base model, devel-oped using extensive phase 1 and 2 data, were used to develop the full covariate model,which was fit to sparse phase 3 data.

ResultsVedolizumab pharmacokinetics was described by a 2-compartment model withparallel linear and nonlinear elimination. Using reference covariate values, linearelimination half-life of vedolizumab was 25.5 days; linear clearance (CLL) was0.159 L/day for UC and 0.155 L/day for CD; central compartment volume of dis-tribution (Vc) was 3.19 L; and peripheral compartment volume of distribution was1.66 L. Interindividual variabilities (%CV) were 35% for CLL and 19% for Vc;residual variance was 24%. Only extreme albumin and body weight values wereidentified as potential clinically important predictors of CLL.

ConclusionsPopulation pharmacokinetic parameters were similar in patients with moderatelyto severely active UC and CD. This analysis supports use of vedolizumab fixeddosing in these patients. Clinicaltrials.gov Identifiers: NCT01177228; NCT00783718(GEMINI 1); NCT00783692 (GEMINI 2); NCT01224171 (GEMINI 3).

Aliment Pharmacol Ther 2015; 42: 188–202

ª 2015 Takeda Pharmaceuticals International Co published by John Wiley & Sons Ltd.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use,

distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.doi:10.1111/apt.13243

188

Alimentary Pharmacology and Therapeutics

http://creativecommons.org/licenses/by-nc/4.0/

INTRODUCTIONVedolizumab, a humanised monoclonal antibody thatbinds specifically to the a4b7 integrin, is indicated for thetreatment of patients with moderately to severely activeulcerative colitis (UC) and Crohn’s disease (CD).1 Bybinding to cell surface-expressed a4b7, vedolizumabblocks the interaction of a subset of memory gut-homingT lymphocytes with mucosal addressin cell adhesionmolecule-1 (MAdCAM–1) expressed on endothelial cells.Consequently, migration of these cells into inflamedintestinal tissue is inhibited.1 The specificity of ve-dolizumab results in a novel gut-selective mechanism ofaction that differs from that of other currently marketedbiologic agents for the treatment for UC and CD, includ-ing natalizumab and tumour necrosis factor-a (TNF-a)antagonists.

The pharmacokinetics of other therapeutic monoclo-nal antibodies used for the treatment of UC and CDhave been reported previously.2 Several factors are asso-ciated with accelerated clearance of these antibodies,including the presence of anti-drug antibodies (ADAs),sex, body size, concomitant immunosuppressant use,inflammatory bowel disease (IBD) type, albumin concen-tration and degree of systemic inflammation.3 Further-more, a consistent relationship between efficacy andexposure, in distinction to drug dose, has been observedfor many of these agents; that is, higher trough drugconcentrations are associated with greater efficacy.4 Dif-ferences in drug clearance may be an important explana-tion for this observation. Therefore, a betterunderstanding of the determinants of clearance for thera-peutic antibodies may result in optimization of drug dos-ing regimens.

The single-dose pharmacokinetics, pharmacodynamics(a4b7 receptor saturation), safety and tolerability of ve-dolizumab have been investigated over a dose range of0.2–10 mg/kg in healthy volunteers (intravenous [IV]infusion) (Takeda Pharmaceuticals International, Inc.,data on file). After reaching peak concentrations, ve-dolizumab serum concentrations fell in a generally biex-ponential fashion until concentrations reachedapproximately 1–10 ng/mL. Thereafter, concentrationsappeared to fall in a nonlinear fashion. The multiple-dose pharmacokinetics and pharmacodynamics of ve-dolizumab have been investigated following IV infusionsof 0.5 and 2 mg/kg in patients with CD5 and patientswith UC6 and following IV infusions of 2, 6, and 10 mg/kg in patients with UC.7 Vedolizumab pharmacokineticswas generally linear following IV infusion over the dose

range of 2–10 mg/kg in patients with UC.7 After multi-ple-dose administration, rapid and near complete a4b7receptor saturation was achieved following the first doseof vedolizumab in patients with UC.7

The efficacy and safety of vedolizumab 300 mg IVinduction therapy and vedolizumab 300 mg IV mainte-nance therapy administered every 8 weeks (Q8W) orevery 4 weeks (Q4W) were demonstrated in patientswith moderately to severely active UC in the GEMINI 1trial and in patients with moderately to severely activeCD in the GEMINI 2 and 3 trials.8–10 The exposure-response (efficacy) relationships of vedolizumab induc-tion and maintenance therapy in these patients havebeen presented elsewhere.8, 9, 11, 12

Here, we report a comprehensive population pharma-cokinetic and pharmacodynamic analysis of vedolizumabtherapy in patients with UC and CD. Our objectiveswere to (i) characterise the pharmacokinetics of ve-dolizumab in patients who received repeated IV infu-sions of vedolizumab 300 mg for up to 52 weeks; (ii)identify clinically relevant determinants of vedolizumabclearance in patients; and (iii) describe the pharmacoki-netic–pharmacodynamic relationship of vedolizumab inpatients using MAdCAM-1 as the pharmacodynamicendpoint.

MATERIALS AND METHODS

Study design and sample collectionAnalyses were conducted using vedolizumab serum con-centrations obtained from 5 randomised, placebo-con-trolled clinical studies: a phase 1 study in healthyvolunteers, a phase 2 study in patients with active UC(NCT01177228), a phase 3 study in patients with moder-ately to severely active UC [GEMINI 1 (NCT00783718)],and 2 phase 3 studies in patients with moderately toseverely active CD [GEMINI 2 (NCT00783692) andGEMINI 3 (NCT01224171)] (Table S1). The studydesigns and clinical data for the phase 2 and 3 studieshave been previously reported.7–10 All study protocolsand consent forms were approved by institutional reviewboards or ethics committees at the study sites, and stud-ies were conducted in accordance with the principles ofgood clinical practice and the Declaration of Helsinki.All patients provided written informed consent beforestudy participation.

Details of the blood sampling times for pharmacoki-netic, pharmacodynamic (MAdCAM-1), and ADA analy-ses are provided in Table S1. Extensive pharmacokinetic

Aliment Pharmacol Ther 2015; 42: 188–202 189

ª 2015 Takeda Pharmaceuticals International Co published by John Wiley & Sons Ltd.

Population pharmacokinetics-pharmacodynamics of vedolizumab

and pharmacodynamic sampling was used in the phase 1and 2 studies, whereas sparse sampling was used in thephase 3 studies. Pharmacodynamic samples were not col-lected in GEMINI 3.

AssaysVedolizumab serum concentrations were determinedusing a sandwich enzyme-linked immunosorbent assay(ELISA), with a lower limit of quantification of 1.25 ng/mL at a 1:100 dilution (125 ng/mL in undiluted serum).The upper limit of quantification was 8 lg/mL; serumsamples with vedolizumab concentrations greater than8 lg/mL were diluted to within the assay range. Theassay was validated at Quest Pharmaceutical Services(Newark, DE, USA). The accuracy of the assay rangedfrom �2.5% to 10.1% difference, intra-sample precisionranged from 1.8% to 3.1% CV, and inter-sample preci-sion ranged from 4.0% to 16.2% CV.

To quantitate a4b7 integrin saturation by vedolizumabin peripheral blood, a MAdCAM–1–Fc binding interfer-ence flow cytometry assay was developed. In this phar-macodynamic assay, inhibition of MAdCAM-1-Fcbinding to a4b7-expressing peripheral blood cells by ve-dolizumab in the blood is used as a measure of theextent of a4b7 saturation by vedolizumab.1 The assay,which was developed by Millennium Pharmaceuticals,Inc. (d/b/a Takeda Pharmaceuticals International, Co.)and validated at Esoterix Center for Clinical TrialResearch (Brentwood, TN, USA), demonstrated an over-all intra-sample variability of 6% CV and an intra-sub-ject variability of 20% CV.

The presence of ADAs was determined using a vali-dated, biotinylated, bridging ELISA and 2 dilutions ofserum (1:5 and 1:50). All samples that screened positivewere further diluted to determine the final ADA titre usingstandard techniques. If both screening dilutions were neg-ative, the sample was considered negative. Patients wereclassified as positive for ADAs if antibodies were detectedat any visit; otherwise, they were classified as negative.

Data assemblyThe dosing, covariate and pharmacokinetic-pharmacody-namic data were merged and formatted for the popula-tion analysis using R, Version 2.10.1 or higher (http://www.r-project.org/). Vedolizumab serum concentrationmeasurements that were missing, or any values withunknown or missing associated observation times, dosetimes, dose amounts or dosing intervals, were excludedfrom the analysis. MAdCAM-1 measurements were trea-ted similarly. All samples with vedolizumab concentra-

tions below the limit of quantification (BLQ) (n = 3189)were not evaluated during the population pharmacoki-netic model development. More than half of the BLQobservations (n = 1722) were samples obtained prior tothe first vedolizumab dose.

Covariates present in the population pharmacokineticdata set were serum C-reactive protein (CRP), serum albu-min, faecal calprotectin, body weight, disease activity[Crohn’s Disease Activity Index (CDAI), complete Mayoscore, partial Mayo score], Mayo endoscopic subscore,age, sex, ADA status (positive or negative), prior TNF-aantagonist therapy status (na€ıve or failed), body massindex (BMI), serum globulin, IBD diagnosis (CD or UC),lymphocyte count and concomitant therapy use (metho-trexate, azathioprine, mercaptopurine or aminosalicy-lates). The start date and end date of concomitant therapywere populated in the data set to evaluate the time-depen-dent effects of concomitant treatments. Covariates withmissing data were imputed using different imputationmethods, based on the remaining available data (e.g. med-ian of the remaining values). No covariates present in thedata set were missing more than 10% of values.

Population pharmacokinetic model developmentThe population pharmacokinetic analysis for repeatedmeasures was conducted using a nonlinear mixed effectsmodelling approach (NONMEM 7, Version 7.2; ICONDevelopment Solutions, Hanover, MD, USA).13 The basepopulation pharmacokinetic model was developed usingthe first-order conditional estimation with g-e interaction(FOCEI) method and extensively sampled phase 1 and 2data. Results from the base model were subsequently usedas prior information to selectively inform a subset of pop-ulation pharmacokinetic model parameters in the full co-variate model, which was fit to sparse phase 3 data fromGEMINI 1, 2 and 3 using the full Bayesian Markov ChainMonte Carlo (MCMC) method. All parameter estimateswere reported with Bayesian 95% credible intervals (CDIs)as a measure of estimation uncertainty.

A covariate modelling approach emphasising parameterestimation rather than stepwise hypothesis testing wasimplemented for the population pharmacokinetic analy-sis.14 First, predefined covariate-parameter relationshipswere identified based on exploratory graphics, scientificinterest, and mechanistic plausibility. Then a full covariatemodel was constructed with care to avoid correlation orcollinearity in predictors; covariates with correlation coef-ficients greater than approximately 0.35 were not simulta-neously included as potential predictors. Construction ofthe full model was also guided by evaluating the adequacy

190 Aliment Pharmacol Ther 2015; 42: 188–202


M. Rosario et al.

http://www.r-project.org/

http://www.r-project.org/

of the study design and covariate data to support quantifi-cation of the covariate effects of interest.

During development of the covariate model, strongcorrelations were identified between the following covari-ates: body weight–BMI, sex–body weight, CRP–albumin,CRP–faecal calprotectin, CRP–globulin, albumin–globu-lin, complete Mayo score–partial Mayo score, Mayoendoscopic subscore–complete Mayo score, and Mayoendoscopic subscore–partial Mayo score. Therefore, sex,CRP, complete Mayo score, Mayo endoscopic subscore,globulin, and BMI were excluded from the full covariatemodel. As the effects of sex, CRP, and Mayo endoscopicsubscore on the pharmacokinetics of vedolizumab couldnot be uniquely estimated in the full model given theircorrelation with other covariates, any remaining effectsof these covariates were independently evaluated in anexploratory post hoc fashion once the population phar-macokinetic model was finalised.

Body weight was chosen to represent changes in ve-dolizumab pharmacokinetics as a function of body sizeand was described using an allometric model with a refer-ence weight of 70 kg. The other continuous covariates ofalbumin, faecal calprotectin, partial Mayo score, age, andCDAI score entered the model as power functions norma-lised by a reference value. The categorical covariates ofprior TNF-a antagonist therapy status, ADA status, con-comitant therapy use, and IBD diagnosis entered themodel as power functions, with a separate dichotomous(0, 1) covariate serving as an on-off switch for each effect.Time-dependent covariates were body weight, albumin,faecal calprotectin, and concomitant therapy use. Theeffect of IBD diagnosis on linear clearance (CLL) wasinvestigated by modelling separate CLL parameters forpatients with UC and those with CD, while the effect ofIBD diagnosis on central compartment volume of distri-bution (Vc) was evaluated by including diagnosis as a pre-dictor of Vc in the covariate model.

Inferences about the clinical relevance of parameterswere based on the resulting parameter estimates andmeasures of estimation precision (Bayesian 95% CDIs)from the full model. In the absence of an exposure-response relationship for efficacy-related clinical end-points to provide a context for interpretation of pharma-cokinetic variability, covariate effect sizes on CLL greaterthan � 25% of the normalised reference value were pro-posed as clinically meaningful changes.

Further details regarding the population pharmacoki-netic and pharmacodynamic analysis methods, includingmodelling assumptions and model evaluation, are pro-vided in Appendix S1.

Population pharmacokinetic-pharmacodynamic modeldevelopmentThe population pharmacokinetic-pharmacodynamicanalysis for repeated measures was conducted using anonlinear mixed effects modelling approach (NONMEM,Version 7.2).13 The pharmacokinetic-pharmacodynamicdata were modelled using a sequential approach, whereindividual predicted vedolizumab serum concentrationsfrom the population pharmacokinetic model were usedto drive the pharmacodynamic response. The pharmaco-dynamic evaluations were based on percentage of MAd-CAM-1 binding by lymphocytes expressing high levels ofa4b7 integrin (CD4+ CD45ROhigh).

A direct effect sigmoid Emax model was chosen asthe structural model to describe the pharmacokinetic-pharmacodynamic relationship of vedolizumab asfollows:

MAdCAM-1 ¼ E0 � 1� Emax � Concc

ECc50 þ Concc

� �

where E0 is the baseline MAdCAM-1 percent binding,Emax is the maximum effect, Conc is the vedolizumabserum concentration, EC50 is the vedolizumab serumconcentration at half-maximum effect, and c is the Hill-coefficient or slope factor. No formal covariate modellingor model evaluation was conducted.

RESULTS

Pharmacokinetic analysis data setThe pharmacokinetic analysis data set consisted of 2554individuals who contributed 18 427 evaluable ve-dolizumab serum samples, including 87 healthy volun-teers from the phase 1 study, 46 patients from the phase2 study (UC), and 891, 1115, and 415 patients from thephase 3 GEMINI 1 (UC), GEMINI 2 (CD), and GEMINI3 (CD) studies, respectively.

Demographics and other characteristics of the pharma-cokinetic analysis data set are summarised in Table 1. Thedata set consisted of 1290 men and 1264 women with agesranging from 18 to 78 years and baseline body weightsranging from 28 to 170 kg. A total of 1530 individuals hadCD and 937 had UC; 87 were healthy volunteers.

The median (interquartile range) vedolizumab troughserum concentration-time profiles for patients with UCfrom GEMINI 1 and patients with CD from GEMINI 2are shown in Figure 1.




Population pharmacokinetic modelling resultsBase pharmacokinetic model. Vedolizumab pharmacoki-netics was described by a 2–compartment model withparallel linear and nonlinear elimination. A 2-compart-ment model resulted in a significant improvement ingoodness-of-fit criteria over a 1–compartment model, asdid a parallel linear and nonlinear elimination model overa linear model. The population pharmacokinetic model ofvedolizumab is represented diagrammatically in Figure 2.

Final pharmacokinetic model. Pharmacokinetic parame-ter estimates and 95% CDIs from the final populationpharmacokinetic model are shown in Table 2, and theirinterindividual variability estimates are shown in Table

S2. The parameter estimates and overlapping 95% CDIsindicated that vedolizumab CLL was the same in patientswith UC (0.159 L/day) and those with CD (0.155 L/day).Individual CLL estimates were distributed over a widerange as represented in Figure 3. The half-life of ve-dolizumab for the linear elimination phase was25.5 days; individual estimates ranged from 14.6 to36.0 days (5th and 95th percentiles, respectively). Inter-individual variability estimates (% CV) from the modelwere 34.6% for CLL and 19.1% for Vc, indicative of mod-erate to large unexplained variability (Table S2).

The residual error (unexplained random residual vari-ability in the model) was 23.5% (% CV), which is con-sidered relatively small (Table 2). Standard deviationsand shrinkage estimates of interindividual random effectsare presented in Table S3. Goodness-of-fit plots from thefinal population pharmacokinetic model are presented inFigures 4 and S1. These plots indicate that the full covar-iate pharmacokinetic model was consistent with theobserved data and no systematic bias was evident.

Table 1 | Summary of demographics and othercharacteristics of the pharmacokinetic analysis data set(N = 2554)

Categorical covariate n (%)

SexWomen 1264 (50)Men 1290 (50)

Disease diagnosisCrohn’s disease 1530 (60)Ulcerative colitis 937 (37)Healthy volunteers 87 (3)

Mayo Endoscopic Subscore1 1 (0.039)2 408 (16)3 482 (19)Missing* 1663 (65)

Prior TNF-a antagonist therapy statusFailed 1321 (52)Na€ıve 1100 (43)Missing† 133 (5)

ADA statusPositive (≥1 positive titre) 124 (5)Negative (no positive titres) 2430 (95)

Continuous covariate n Median (range)

Age, years 2554 36 (18–78)Body weight, kg 2554 68 (28–170)Albumin, g/L 2467 37 (11–53)C-reactive protein, mg/L 1576 11 (0.2–200)Faecal calprotectin, mg/kg 2421 720 (23.75–20 000)CDAI score 1530 320 (93–580)Mayo Score 891 9 (3–12)Partial Mayo Score 937 6 (1–9)

ADA, anti-drug antibody; CDAI, Crohn’s Disease ActivityIndex; TNF-a, tumour necrosis factor-a.

* Data not collected in phase 1 and 2 studies or in GEMINI 2and 3 studies.

† Data not collected in phase 1 and 2 studies.

Crohn’s disease

6

10.0

1.0

0.1

10.0

1.0

Obs

erve

d ve

doliz

umab

trou

gh c

once

ntra

tion

(µg/

mL)

0.1

12 18 24

Ulcerative colitis

30 36 42 48

6 12 18 24 30

Time (week)

Placebo Vedolizumab Q4W Vedolizumab Q8W

36 42 48

Figure 1 | Median (interquartile range) of observedvedolizumab trough serum concentration vs. nominalsampling time in patients with UC (GEMINI 1) andpatients with CD (GEMINI 2) during maintenancetreatment with placebo or vedolizumab 300 mg every4 weeks (Q4W) or every 8 weeks (Q8W). All patients(including those in the placebo group) received 2 dosesof vedolizumab 300 mg during induction (at weeks 0and 2). The median value is shown as a point and theinterquartile range is represented by a vertical bar.



M. Rosario et al.

The relative changes in CLL for a reference patientwith various covariate values are illustrated in Figure 5.The impact of these covariates on CLL was evaluated

univariately. The point estimates and 95% CDIs for theeffects of covariates on CLL are presented in Table S4. Ingeneral, the 95% CDIs were narrow indicating that theeffect of each covariate on vedolizumab CLL was welldefined. Only the effects of albumin and body weight atextreme values had the potential to be clinically mean-ingful (effect sizes greater than �25%). The CLL valuesfor patients with albumin levels of 4.7 and 3.2 g/dL wereapproximately 0.8 and 1.3 times, respectively, that of thereference patient (albumin, 4 g/dL) (Figure 5). The CLLvalues for patients of 40 and 120 kg were approximately0.8 and 1.2 times, respectively, that of the referencepatient (weight, 70 kg) (Figure 5). A patient of 120 kgwith a serum albumin concentration of 4.0 g/dL had a19% probability of having a CLL value greater than thepre-specified criterion for clinical significance. Whenevaluated across a representative range of covariate val-ues and categories, the effects of faecal calprotectin,CDAI score, partial Mayo score, age, prior TNF–a antag-onist therapy status, ADA status, and concomitant ther-apy use on vedolizumab CLL were not consideredclinically relevant as the covariate effect sizes were lessthan �25% from the reference values (Figure 5 andTable S4). In addition, the 95% CDIs for these covariateeffects contained the null effect value.

The final population pharmacokinetic model wasrerun with all covariate effects and pharmacokineticparameters fixed to estimates from the final model (in-terindividual variances were re-estimated), and anyremaining effects of sex on CLL and Vc were quantified.The results of this analysis suggest that, after adjustmentfor other predictors of vedolizumab CLL, the CLL and Vc

values were approximately 10% lower and 6% lower,respectively, for a female patient compared with a malepatient. However, these effects were not considered clini-cally relevant as the covariate effect sizes were less than� 25% from the reference values (male patient).

Central compartment

Peripheral compartment

Intercompartmental clearance

Dose

Nonlinear clearance(e.g. target-mediated elimination)

Linear clearance(CLL)

CLNL =Vmax

Km+ Conc

Figure 2 | Diagrammaticrepresentation of thepopulation pharmacokineticmodel of vedolizumab. Conc,vedolizumab concentration;Km, concentration at half-maximum elimination rate;Vmax, maximum eliminationrate.

Table 2 | Parameter estimates from the finalpopulation pharmacokinetic model for vedolizumab

Parameter Estimate*Bayesian 95%CDI

Ulcerative colitis: CLL 0.159 L/day 0.153–0.165Crohn’s Disease: CLL 0.155 L/day 0.149–0.161Central compartmentvolume ofdistribution (Vc)

3.19 L 3.14–3.25

Peripheral compartmentvolume ofdistribution (Vp)

1.65 L 1.59–1.71

Intercompartmentalclearance (Q)

0.12 L/day 0.112–0.129

Maximum eliminationrate (Vmax)

0.265 mg/day 0.219–0.318

Concentration athalf-maximumelimination rate (Km)

0.964 lg/mL 0.706–1.27

Proportional residualerror variance (r2

prop)0.0554(% CV = 23.5)

0.0539–0.0568

CDI, credible interval; CLL, clearance of linear elimination path-way; CV, coefficient of variation.

* Parameter estimate and 95% credible interval were derivedfrom the median and 2.5th and 97.5th percentiles, respectively,of the Bayesian posterior probability distributions from 4 Mar-kov Chain Monte Carlo chains. Separate values of CLL weremodelled for patients with UC and those with CD with a sharedinterindividual variance term and shared covariate effectsexcept for partial Mayo score and CDAI score. The referenceindividual weighs 70 kg; is 40 years old; has an albumin level of4 g/dL, a fecal calprotectin level of 700 mg/kg, a CDAI scoreof 300 (for patient with CD), a partial Mayo score of 6 (forpatient with UC), a diagnosis of UC (for Vc parameter), and noconcomitant therapy use; and is anti-drug antibody negativeand tumour necrosis factor-a antagonist therapy na€ıve.




Addition of the sex effect explained approximately 4.2%and 6.0% of the unexplained interindividual variability inCLL and Vc, respectively.

The final population pharmacokinetic model also wasre-run to estimate any remaining effect of CRP on CLL.The results suggest that, after adjustment for other pre-dictors of vedolizumab CLL (such as albumin), the effectof CRP on CLL was not clinically relevant as the covari-ate effect size was less than � 25% from the referencevalue (CRP, 11 mg/dL). The addition of the CRP effectexplained <1% of the unexplained interindividual vari-ability in CLL.

For patients with UC, the effect of Mayo endoscopicsubscore was evaluated graphically by plotting individualCLL estimates from the final population pharmacokineticmodel by endoscopic subscore at week 6 (Figure 6).From this analysis, at week 6 (end of inductiontreatment), patients with an endoscopic subscore of 3had on average 25% higher CLL than patients with anendoscopic subscore of 0.

Pharmacokinetic model evaluation. The final populationpharmacokinetic model and parameter estimates wereevaluated with a predictive check method and Bayesian95% CDIs derived from the posterior probability distri-butions. The basic premise of a predictive check is that a

model and parameters derived from an observed data setshould produce simulated data that are similar to theoriginal observed data. The predictive check plots dem-onstrated overall good agreement between the observedand simulated data (Figures S2–S4). The precision of theparameter estimates was assessed by evaluating theBayesian 95% CDIs (Tables 2, S2, and S4). Overall, thestructural pharmacokinetic model parameters, covariateeffects and variance parameters were estimated withgood precision.

Pharmacokinetic-pharmacodynamic data setThe vedolizumab population pharmacokinetic-pharmaco-dynamic data set was composed of 593 individuals

30

Crohn’s disease25

20

15

Per

cent

age

of p

atie

nts

10

5

0

0.0 0.1 0.2 0.3CLL (L/day)

0.4 0.5 0.6 0.7

Ulcerative colitis

Figure 3 | Distribution of individual vedolizumab linearclearance (CLL) estimates from the final populationpharmacokinetic model in patients with UC andpatients with CD.

500

500

400

400

300

300

Obs

erve

d ve

doliz

umab

conc

entr

atio

n (µ

g/m

L)

Population predicted vedolizumab concentration (µg/mL)

200

200

100

1000

0

500400300Individual predicted

vedolizumab concentration (µg/mL)

2001000

500

400

300

Obs

erve

d ve

doliz

umab

conc

entr

atio

n (µ

g/m

L)200

100

0

Figure 4 | Goodness-of-fit plots: observed vedolizumabserum concentration vs. population and individualpredicted vedolizumab concentration from the finalpopulation pharmacokinetic model. Values are shownas points with a dashed pink loess trend line throughthe data. A line of identity (solid black) is shown forreference.



M. Rosario et al.

contributing a total of 2442 evaluable MAdCAM-1observations. The data set consisted of 297 patients withUC and 296 patients with CD (from the phase 2 studyand phase 3 GEMINI 1 and 2 studies).

During the analysis, the log-transformed values ofMAdCAM-1 (free a4b7 receptors not blocked by ve-dolizumab) were modelled. A plot of observed MAd-CAM-1 measurements (percentage of free a4b7receptors) vs. observed vedolizumab serum concentra-tions for patients with UC and CD from GEMINI 1 and2, respectively, is presented in Figure 7. Plots of observedMAdCAM-1 measurements vs. time by treatment regi-men for patients with UC and CD in GEMINI 1 and 2,respectively, are presented in Figure 8. The percentage offree a4b7 receptors declined rapidly after the first doseand this reduction was maintained during repeated IVinfusions of vedolizumab 300 mg.

Population pharmacokinetic-pharmacodynamicmodelling resultsStructural parameter estimates from the population phar-macokinetic-pharmacodynamic model are presented inTable 3; parameters were estimated with adequate preci-sion. An attempt was made to model a placebo effect,but the estimated effect was negligible. The lack of anapparent placebo effect was consistent with the observedMAdCAM-1 data in placebo–treated patients.

The base model provided a reasonable description ofthe data as judged by visual inspection of diagnosticplots (Figure S5), but some deficiencies in the modelwere noted. Variance parameter estimates were indicativeof moderate to large unexplained interindividual andresidual variability, with estimates of 41.8% CV for E0,0.551 (SD logistic distribution) for Emax, and 78.3% CVfor the exponential residual error variance (r2

exp)(Table 3).

DISCUSSIONWe developed a population pharmacokinetic model forvedolizumab administered IV to healthy volunteers andpatients with moderately to severely active UC and CD,using data collected in clinical trials with identicaldesigns and sampling schedules. These design featuresallowed the direct comparison of the disposition andpharmacokinetic variability in vedolizumab in patientswith UC and patients with CD. The estimated half-lifeof vedolizumab was not different between the 2 diseasesand was 25.5 days for the reference patient [individualestimates ranged from 14.6 to 36.0 days (5th and 95thpercentiles respectively)], which is typical of human IgG1

(25 days) and of monoclonal antibodies of the IgG1 sub-class. This half-life is longer than values for other cur-rently marketed biologic treatments for UC and CD.Clinicians should be aware of the relatively long half-life

4.7

4.2

3.7

3.2

2.7

120

100

80

60

40

Aminosalicylates

Wei

ght (

kg)

Alb

umin

(g/

dL)

Methotrexate

Mercaptopurine

Azathioprine

REFERENCE

0.50 0.75 1.00Normalised vedolizumab CLL

1.25 1.50 1.75

Figure 5 | Effect of covariates on vedolizumab linearclearance (CLL). Each point and line represent themedian and 95% credible interval, respectively, of theBayesian posterior distribution of normalised samplesof vedolizumab CLL adjusted for the covariate. Thereference individual weighs 70 kg; is 40 years old; hasan albumin level of 4 g/dL, a faecal calprotectin levelof 700 mg/kg, a CDAI score of 300 (for patient withCD), a partial Mayo score of 6 (for patient with UC),and no concomitant therapy use; and is ADA negativeand TNF-a antagonist therapy na€ıve. Albumin: 2.7, 3.2,3.7, 4.2 and 4.7 g/dL represent the 6th, 18th, 70th,85th, and 98.5th percentiles, respectively, of baselinealbumin levels for patients in GEMINI 1, 2, and 3.Weight: 40, 60, 80, 100, and 120 kg represent the1.5th, 30th, 71st, 92nd and 98th percentiles,respectively, of baseline weight values for patients inGEMINI 1, 2, and 3. The vertical black line is drawn atthe reference point estimate, and the shaded region is� 25% of the reference point estimate chosen torepresent an uncertainty range of clinicalunimportance.




of vedolizumab in circumstances when it is desirable tominimise exposure to the drug.

A 2-compartment pharmacokinetic model consistingof parallel clearance via a nonlinear pathway (CLNL) anda linear pathway (CLL) from the central compartmentwas selected as the base model for vedolizumab. Thenonlinear elimination was best described by Michaelis–Menten elimination.15 Physiologically, the nonlinearpathway is thought to be due to clearance by saturable,target-mediated mechanisms such as receptor-mediatedendocytosis. In contrast, the linear pathway representscomponents that are nonsaturable at therapeutic concen-trations, such as Fc-mediated elimination. Parallel elimi-nation is typical of monoclonal antibodies withdisposition that is affected by binding to the target, inthe case of vedolizumab, the a4b7 integrin on circulatingT lymphocytes.16 Similar target-mediated drug disposi-tion properties have been reported for efalizumab, toc-ilizumab and cetuximab.17–19 In contrast, the eliminationof TNF-a antagonists, such as infliximab and goli-mumab, was best described by a single linear eliminationpathway.20–22 TNF-a exists in both soluble and mem-brane-bound forms and is present in abnormally highconcentrations in serum and gut mucosa in patients withIBD. The localization of TNF-a in inflammatory tissues

may make it difficult to rapidly achieve target saturationat therapeutic concentrations because of slow redistribu-tion of the drug from plasma to the target sites.23 Fur-thermore, variability in TNF-a concentrations indifferent compartments results in drug redistribution,with possible effects on pharmacokinetic-efficacy/safetyrelationships of TNF-a antagonists.24 The clinical impli-cations of these differences in elimination pathways arenot currently understood.

The vedolizumab CLL values estimated from the phar-macokinetic model are consistent with those of othermonoclonal antibodies that are administered intrave-nously.18, 20, 22 Although some authors have reportedthat the clearance of monoclonal antibodies could beaffected by IBD type,3 no apparent differences wereobserved in the CLL of vedolizumab in patients with UCand those with CD. Vedolizumab CLL for a 40-year-old,70-kg patient with a serum albumin concentration of4 g/dL was 0.159 L/day for a patient with UC and0.155 L/day for a patient with CD.

We evaluated the potential effects of intrinsic andextrinsic covariates (body weight, age, albumin, faecalcalprotectin, CDAI score, partial Mayo score, concomi-tant therapy use, ADA status, and prior TNF-a antago-nist therapy status) on vedolizumab CLL. Extreme values

0.6

0.5

0.4

0.3

Ved

oliz

umab

CL L

(L/

day)

0.2

0.1

0.0

0(n = 56)

1(n = 223)

2(n = 230)

Mayo endoscopic subscore

3(n = 199)

Figure 6 | Individualvedolizumab linear clearance(CLL) estimates by Mayoendoscopic subscore at theend of induction (week 6) forpatients with UC (GEMINI 1).Midlines represent medians.Box limits represent 25th and75th percentiles. Whiskers(error bars) represent highestand lowest points within1.5 9 interquartile range.Individual points representoutliers.



M. Rosario et al.

of albumin had important effects on vedolizumab CLL.Specifically, CLL increased as albumin concentrationsdecreased. Albumin concentrations below 3.2 g/dL, for apatient of 70 kg, were associated with increased ve-dolizumab CLL that was greater than the pre-specifiedcriterion for clinical significance. Similar association oflow albumin concentrations with increased clearance hasbeen reported in population pharmacokinetic analyses ofother monoclonal antibodies.21, 25 Although the mecha-nism of this interaction is incompletely understood, theneonatal Fc receptor (FcRn)–which is expressed in awide variety of cells and tissues throughout the body,including the vascular endothelium, monocytes, macro-phages, dendritic cells, hepatocytes and epithelial cells ofthe intestine, renal proximal convoluted tubules, andupper airways–facilitates IgG and albumin homeostasisby salvaging these molecules from proteolysis and recy-cling them into the central circulatory system. It hasbeen postulated that decreased expression and/or activityof FcRn in patients with IBD results in low albumin andIgG concentrations due to less efficient salvage of theseproteins; the net result is increased antibody clearance.26

However, this hypothesis has not been proven. Anotherpossible explanation is that inflammation of the gastroin-testinal tract in patients with IBD can result in anunconventional route of elimination. Specifically, in the

setting of severe colitis, patients may develop a protein-losing colopathy in which large amounts of protein arelost from the luminal surface. Albumin may thereforeserve as a surrogate marker for loss of endogenous IgGand monoclonal antibodies via this pathway. Thishypothesis is in part supported by observations inpatients with severe colitis treated with infliximab, inwhom concentrations of the monoclonal antibody werehigh in the stool and low in serum; successful treatmentwas associated with resolution of this phenomenon.27

Vedolizumab concentrations in faeces were not measuredduring the GEMINI trials, precluding us from evaluatingthis hypothesis.

The second important covariate identified was bodyweight, which was positively correlated with vedolizumabCLL. A patient of 120 kg with a serum albumin concen-tration of 4.0 g/dL had a 19% probability of having CLLgreater than the pre-specified criterion for clinical signifi-cance. Measures of body size are the most commonlyidentified covariates influencing the pharmacokinetics oftherapeutic monoclonal antibodies.25 The impact of bodyweight on vedolizumab CLL is consistent with thatreported in population pharmacokinetic analyses of othertherapeutic monoclonal antibodies.25

Other potential covariates had lesser effects onvedolizumab CLL than weight and albumin in the

60

50

40

30

Obs

erve

d M

AdC

AM

-1 (

%)

Observed vedolizumab concentration (µg/mL)

20

10

0

0 50 100 150 200 250 300

Figure 7 | Receptor (a4b7)saturation plot: observedMAdCAM-1 vs. observedvedolizumab serumconcentration in patients withUC (GEMINI 1) and patientswith CD (GEMINI 2). Baselinedata are included.




current analysis. Patients with UC who had lower Mayoendoscopic subscores had, on average, lower ve-dolizumab CLL and, therefore, higher serum concentra-tions than patients who had higher endoscopicsubscores. This finding is consistent with what has beenreported for TNF-a antagonists.4, 28–30 However, theseresults should be interpreted with caution as it is possi-ble that the relationship between vedolizumab CLL andendoscopic subscore is not causal but merely reflects theassociation between drug-losing enteropathy and muco-sal healing.

Intensity of inflammation has been reported as a posi-tive predictor of clearance for other monoclonal antibod-ies.3 A post hoc exploratory analysis revealed that, afteraccounting for the effects of other covariates (such asalbumin) in the existing pharmacokinetic model, theremaining effect of CRP on vedolizumab CLL is not clin-ically relevant and explained less than 1% of the unex-plained interindividual variability in CLL. Therefore, as

albumin and CRP were strongly correlated, any potentialeffect of CRP on vedolizumab CLL was alreadyaccounted for in the model by incorporating albumin.Similar results have been reported recently by Wade atal. for certolizumab pegol in patients with CD.31 In con-trast, high CRP concentrations were strongly associatedwith increased clearance of anti-TNF-a monoclonal anti-bodies in the literature; however, effects of covariatessuch as albumin that appear to be correlated withinflammatory markers in patients with IBD were notinvestigated in these analyses.3

The development of ADAs has been reported toincrease infliximab clearance.21 In the current analysis,the presence of ADAs was estimated to increase ve-dolizumab CLL by only 12%. Inferences regarding thisimpact are limited by the low incidence of ADAsobserved in the GEMINI trials.8–10 In the few patientswho were persistently positive for ADAs in these studies,vedolizumab trough concentrations were below the limit

60Placebo Vedolizumab Q4W

Vedolizumab Q8WVedolizumab to placeboafter 6 week induction

50

40

30

20

10

00 6 12 18 24 30 36 42 48 54 60

60

50

40

30

20

10

00 6 12 18 24 30 36 42 48 54 60

Time (week)

Crohn’s disease Ulcerative colitis

60

50

40

30

20

10

00 6 12 18 24 30 36 42 48 54 60

60

50

40

30

20

10

00 6 12 18 24 30 36 42 48 54 60

Obs

erve

d M

AdC

AM

-1 (

%)

Figure 8 | Observed MAdCAM-1 vs. time for patients with UC (GEMINI 1) and patients with CD (GEMINI 2). Upperleft panel: patients received placebo during induction and maintenance; upper right panel: patients received 2 doses ofvedolizumab 300 mg during induction and vedolizumab 300 mg every 4 weeks (Q4W) during maintenance; lowerleft panel: patients received 2 doses of vedolizumab 300 mg during induction and vedolizumab 300 mg every8 weeks (Q8W) during maintenance; lower right panel: patients received 2 doses of vedolizumab 300 mg duringinduction and placebo during maintenance.



M. Rosario et al.

of quantification. We believe that sensitization of patientsto monoclonal antibodies is an important cause of treat-ment failure and that vedolizumab is not unique in thisregard.

The current analysis showed no clinically meaningfulimpact of prior TNF-a antagonist therapy status, con-comitant medication use, age (from 18 to 78 years old),and disease activity on vedolizumab CLL. The lack ofassociation between prior TNF-a antagonist therapy sta-tus and vedolizumab CLL is interesting, especially giventhat lack of prior TNF-a antagonist therapy use wasassociated with a higher probability of clinical remissionor response in vedolizumab-treated patients with UCand those with CD.11, 12 Taken together, these observa-tions suggest that the impact of prior TNF-a antagonisttherapy use on vedolizumab efficacy is not related to anyeffect on vedolizumab CLL.

The lack of an effect of thiopurines and methotrexateon vedolizumab CLL differs from effects seen on TNF-aantagonists in patients with UC and CD and in patientswith rheumatoid arthritis, where co-administration ofthese agents is associated with higher trough concentra-tions and lower clearance of TNF-a antagonists.32 Themechanism by which antimetabolites increase concentra-tions of biologic drugs is not well understood; however,modulation of Fcc receptor expression is one possibleexplanation. For example, methotrexate is known todown-regulate Fcc receptors on monocytes and otherFc-receptor subtypes. Another possible explanation isthat the prevention of ADA development could increasedrug exposure by reducing immune-mediated drug clear-ance. The reason behind the lack of effect of concomi-

tant medications on vedolizumab CLL is not currentlyunderstood. A sensitivity analysis was performed todetermine whether the rate of concomitant medicationuse in the vedolizumab population pharmacokinetic dataset was sufficient to achieve at least 80% power to detectno drug interaction, as recommended by the PopulationPharmacokinetic Therapeutic Protein–Drug Interaction(PK TPDI) Working Group.33 This analysis revealed thatthe data set met the sample size requirements to ensureat least 80% power and confirmed that vedolizumab CLLwas not impacted by concomitant use of azathioprine,mercaptopurine, methotrexate or aminosalicylates.34

Interestingly, a4b7 receptor saturation, as measured inthe MAdCAM-1 assay, was maintained at vedolizumabconcentrations considered subtherapeutic, raising thequestion of whether receptor saturation is necessary butnot sufficient for clinical efficacy. The EC50 estimate fromthe population pharmacokinetic-pharmacodynamicmodel was 0.093 lg/mL, suggesting that full saturation isreached at a vedolizumab serum concentration of approx-imately 1 lg/mL. Exposure-efficacy data indicated thatvedolizumab concentrations below 17 and 15 lg/mL atinduction were associated with efficacy similar to placeboin patients with UC and those with CD, respectively.8, 9

This discrepancy might be explained by the fact that theMAdCAM-1 assay measures a4b7 saturation in circulat-ing T-cells, or may be due to a slow onset of action ofthe drug. The MAdCAM-1 assay is insensitive to doseand should not be used for dose selection. Further studiesto evaluate the relative pharmacodynamic contributionsof a4b7 receptor blockade in the peripheral blood vs. inthe tissue compartment are a research priority.

In conclusion, a population model characterising thepharmacokinetic properties of vedolizumab was success-fully developed for patients with UC and CD. The mod-elling results suggested that vedolizumab CLL wassimilar in patients with UC and those with CD. Albu-min and body weight were identified as predictors of ve-dolizumab CLL, but the effects of these covariates wereonly considered clinically meaningful at extreme values.Surprisingly, concomitant use of immunosuppressantshad no clinically relevant impact on vedolizumab CLL, afinding that contrasts with the well–established relation-ship between immunogenicity and TNF-a antagonistconcentrations. This analysis supports use of fixed dos-ing with vedolizumab in patients with UC and thosewith CD.

AUTHORSHIPGuarantor of the article: Maria Rosario.

Table 3 | Parameter estimates from base MAdCAM-1population pharmacokinetic–pharmacodynamic modelfor vedolizumab

Parameter EstimatePercentrelative S.E.

Baseline MAdCAM-1percent binding (E0)

12.1% 3.49

Vedolizumab serumconcentration athalf-maximumeffect (EC50)

0.093 lg/mL 25.8

Maximum effect (Emax) 0.959 0.503Hill-coefficient orslope factor (c)

0.801 11.1

Exponential residualerror variance (r2

exp)0.613(% CV = 78.3)

10.4

CV, coefficient of variation; S.E., standard error.




Author contributions: Irving Fox, Asit Parikh, Ade A. Fa-sanmade, Brian G. Feagan, and William J. Sandborncontributed to the design of the vedolizumab clinical tri-als. Timothy Wyant assisted with the pharmacokinetic,pharmacodynamic and biomarker assays. Maria Rosario,Nathanael L. Dirks, Irving Fox, Brian G. Feagan, WilliamJ. Sandborn and Walter Reinisch provided input into thepharmacokinetic/pharmacodynamic analysis plan.Nathanael L. Dirks and Marc R. Gastonguay performedthe pharmacokinetic/pharmacodynamic analyses. MariaRosario and Nathanael L. Dirks assisted with the draftingof the manuscript.

All authors reviewed and revised the manuscript forimportant intellectual content and approved the finalversion of the manuscript prior to submission.

ACKNOWLEDGEMENTSDeclaration of personal interests: Maria Rosario, AdedigboA. Fasanmade, Timothy Wyant and Irving Fox areemployees of Takeda Pharmaceuticals International Co.and hold equity stake in Takeda. Asit Parikh is anemployee of Takeda Pharmaceuticals International, Inc.and holds equity stake in Takeda. Nathanael L. Dirks andMarc R. Gastonguay are employees of Metrum ResearchGroup, LLC, and were paid consultants to Takeda for thisresearch. William J. Sandborn has received consulting feesfrom Abbott, ActoGeniX NV, AGI Therapeutics Inc., AlbaTherapeutics Corp, Albireo, Alfa Wasserman, Amgen,AM–Pharma BV, Anaphore, Astellas, Athersys Inc.,Atlantic Healthcare Ltd, Aptalis, BioBalance Corp, Boeh-ringer-Ingelheim, Bristol-Myers Squibb, Celgene, CelekPharmaceuticals, Cellerix SL, Cerimon Pharmaceuticals,ChemoCentryx, CoMentis, Cosmo Technologies, Coro-nado Biosciences, Cytokine Pharmasciences, Eagle Phar-maceuticals, EnGene Inc., Eli Lilly, Enteromedics, ExagenDiagnostics Inc., Ferring Pharmaceuticals, Flexio Thera-peutics Inc., Funxional Therapeutics Ltd, Genzyme Corp,Gilead Sciences, Given Imaging, GlaxoSmithKline, HumanGenome Sciences, Ironwood Pharmaceuticals, KaloBiosPharmaceuticals, Lexicon Pharmaceuticals, Lycera Corp,Meda Pharmaceuticals, Merck Research Laboratories,Merck Serono, Millenium Pharmaceuticals, Nisshin Kyo-rin Pharmaceuticals, Novo Nordisk, NPS Pharmaceuticals,Optimer Pharmaceuticals, Orexigen Therapeutics Inc.,PDL Biopharma, Pfizer, Procter and Gamble, PrometheusLaboratories, ProtAb Ltd, Purgenesis Technologies Inc.,Relypsa Inc., Roche, Salient Pharmaceuticals, Salix Phar-maceuticals, Santarus, Schering Plough, Shire Pharmaceu-ticals, Sigmoid Pharma Ltd, Sirtris Pharmaceuticals, SLAPharma UK Ltd, Targacept, Teva Pharmaceuticals, Thera-

kos, Tillotts Pharma AG, TxCell SA, UCB Pharma, ViametPharmaceuticals, Vascular Biogenics Ltd, Warner ChilcottUK Ltd, and Wyeth; research grants from Abbott, Bristol–Myers Squibb, Genentech, GlaxoSmithKline, Janssen, Mi-lennium Pharmaceuticals, Novartis, Pfizer, Procter andGamble, Shire Pharmaceuticals, and UCB Pharma; pay-ments for lectures/speakers bureaux from Abbott, Bristol-Myers Squibb, and Janssen; and holds stock/stock optionsin Enteromedics. Brian G. Feagan has served as a speaker,a consultant, and/or an advisory board member for Ab-bott/AbbVie, Actogenix, Albireo Pharma, Amgen, AstraZeneca, Avaxia Biologics Inc., Avir Pharma, Axcan, BaxterHealthcare Corp., Biogen Idec, Boehringer-Ingelheim,Bristol-Myers Squibb, Calypso Biotech, Celgene, CentocorInc., Elan/Biogen, EnGene, Ferring Pharma, Roche/Genentech, GiCare Pharma, Gilead, Given Imaging Inc.,GlaxoSmithKline, Ironwood Pharma, Janssen Biotech(Centocor), JnJ/Janssen, Kyowa Kakko Kirin Co Ltd, Lexi-con, Lilly, Merck, Millennium, Nektar, Novartis Novonor-disk, Pfizer, Prometheus Therapeutics and Diagnostics,Protagonist, Receptos, Salix Pharma, Serono, Shire, Sig-moid Pharma, Synergy Pharma Inc., Takeda, TevaPharma, TiGenix, Tillotts, UCB Pharma, Vertex Pharma,VHsquared Ltd, Warner-Chilcott, Wyeth, Zealand, andZyngenia; has received research funding from Abbott/AbbVie, Amgen, Astra Zeneca, Bristol-Myers Squibb,Janssen Biotech (Centocor), JnJ/Janssen, Roche/Genen-tech, Millennium, Pfizer, Receptos, Santarus, Sanofi, Till-otts, and UCB Pharma; and is an employee of RobartsResearch Institute, University of Western Ontario. WalterReinisch has served as a speaker, a consultant and/or anadvisory board member for Abbott Laboratories, Abbvie,Aesca, Amgen, AM Pharma, Aptalis, Astellas, Astra Zene-ca, Avaxia, Bioclinica, Biogen IDEC, Bristol-Myers Squibb,Cellerix, Chemocentryx, Celgene, Centocor, Celltrion, Da-none Austria, Elan, Falk Pharma GmbH, Ferring, Galapa-gos, Genentech, Gr€unenthal, Inova, Janssen, Johnson &Johnson, Kyowa Hakko Kirin Pharma, Lipid Therapeutics,MedImmune, Millenium, Mitsubishi Tanabe Pharma Cor-poration, MSD, Novartis, Ocera, Otsuka, PDL, Pharma-cosmos, Pfizer, Procter & Gamble, Prometheus, RobartsClinical Trials, Schering-Plough, Setpointmedical, Shire,Takeda, Therakos, Tigenix, UCB, Vifor, Yakult, Zyngenia,and 4SC; and is an employee of McMaster University andMedical University of Vienna.

Declaration of funding interests: Clinical studies andpopulation pharmacokinetic and pharmacodynamicanalyses were funded by Millennium Pharmaceuticals,Inc. (d/b/a Takeda Pharmaceuticals International, Co.),Cambridge, MA, USA. The population pharmacokinetic



M. Rosario et al.

and pharmacodynamic analyses were performed byNathanael L. Dirks and Marc R. Gastonguay, MetrumResearch Group LLC, Tariffville, CT, USA. Editing assis-tance was provided by Elisabeth R. Wann, PhD, WannMedical Communications, LLC, Wilmette, IL, USA, andwas supported by Takeda Pharmaceuticals International,Inc., Deerfield, IL, USA. The results of these analyseswere previously presented at the 9th Congress of Euro-pean Crohn’s and Colitis Organisation – InflammatoryBowel Diseases 2014 and at the Digestive Disease Week2014 Annual Scientific Meeting. The authors thankMichael Bargfrede, BS, and Sanhita Abrol, MS, for theirassistance with generating the data sets and figures, andDennis Noe, MD, for his technical guidance duringdevelopment of the analysis plan.

SUPPORTING INFORMATIONAdditional Supporting Information may be found in theonline version of this article:Figure S1. Goodness-of-fit plots for the vedolizumab

final population pharmacokinetic model: residual andconditional weighted residual with interaction (CWRESI)vs. time and population predicted vedolizumab concen-tration.

Figure S2. Predictive check for the vedolizumab finalpopulation pharmacokinetic model: induction therapy.Figure S3. Predictive check for the vedolizumab final

population pharmacokinetic model: maintenance therapyevery 4 weeks.Figure S4. Predictive check for the vedolizumab final

population pharmacokinetic model: maintenance therapyevery 8 weeks.Figure S5. Goodness-of-fit plots for the population

pharmacodynamic MAdCAM-1 model.Table S1. Clinical studies included in vedolizumab

population pharmacokinetic–pharmacodynamic analyses.Table S2. Estimates of interindividual variability (%

CV) and correlations from the vedolizumab final popula-tion pharmacokinetic model.Table S3. Standard deviations and shrinkage estimates

of interindividual random effects from the vedolizumabfinal population pharmacokinetic model.Table S4. Covariate parameter estimates from the

vedolizumab final population pharmacokinetic model.Appendix S1. Population pharmacokinetic and

pharmacodynamic analysis methods, including modellingassumptions and model evaluation.

REFERENCES1. Soler D, Chapman T, Yang LL, Wyant

T, Egan R, Fedyk ER. The bindingspecificity and selective antagonism ofvedolizumab, an anti-alpha4beta7integrin therapeutic antibody indevelopment for inflammatory boweldiseases. J Pharmacol Exp Ther 2009;330: 864–75.

2. Paul S, Moreau AC, Del Tedesco E,et al. Pharmacokinetics of adalimumabin inflammatory bowel diseases: asystematic review and meta-analysis.Inflamm Bowel Dis 2014; 20: 1288–95.

3. Ordas I, Mould DR, Feagan BG,Sandborn WJ. Anti-TNF MonoclonalAntibodies in Inflammatory BowelDisease: Pharmacokinetics-BasedDosing Paradigms. Clin Pharmacol Ther2012; 91: 635–46.

4. Colombel JF, Feagan BG, Sandborn WJ,Van Assche G, Robinson AM.Therapeutic drug monitoring ofbiologics for inflammatory boweldisease. Inflamm Bowel Dis 2012a; 18:349–58.

5. Feagan BG, Greenberg GR, Wild G,et al. Treatment of active Crohn’s

disease with MLN0002, a humanizedantibody to the alpha4beta7 integrin.Clin Gastroenterol Hepatol 2008; 6:1370–7.

6. Feagan BG, Greenberg GR, Wild G,et al. Treatment of ulcerative colitiswith a humanized antibody to thealpha4beta7 integrin. N Engl J Med2005; 352: 2499–507.

7. Parikh A, Leach T, Wyant T, et al.Vedolizumab for the treatment of activeulcerative colitis: a randomizedcontrolled phase 2 dose-ranging study.Inflamm Bowel Dis 2012; 18: 1470–9.

8. Feagan BG, Rutgeerts P, Sands BE,et al. Vedolizumab as induction andmaintenance therapy for ulcerativecolitis. N Engl J Med 2013; 369: 699–710.

9. Sandborn WJ, Feagan BG, Rutgeerts P,et al. Vedolizumab as induction andmaintenance therapy for Crohn’sdisease. N Engl J Med. 2013; 369: 711–21.

10. Sands BE, Feagan BG, Rutgeerts P,et al. Effects of vedolizumab inductiontherapy for patients with Crohn’s

disease in whom tumor necrosis factorantagonist treatment failed.Gastroenterology 2014; 147: 618–27.

11. Rosario M, French J, Dirks N, MiltonA, Fox I, Gastonguay M, eds. Exposure-response relationship duringvedolizumab induction therapy inadults with ulcerative colitis. 9thCongress of the European Crohn’s andColitis Organisation; 20–22 February2014; Copenhagen, Denmark: J CrohnsColitis, 2014a.

12. Rosario M, French J, Dirks N, MiltonA, Fox I, Gastonguay M, eds. Exposure-response relationship of vedolizumabafter 6 weeks of treatment in adultswith Crohn’s disease. 9th Congress ofthe European Crohn’s and ColitisOrganisation; 20–22 February 2014;Copenhagen, Denmark: J CrohnsColitis, 2014b.

13. Beal SL, Sheiner LB, Boeckmann AJ,Bauer RJ, eds. NONMEM Users GuidePart I-VIII. Hanover, MD: ICONDevelopment Solutions, 1989–2012.

14. Gastonguay M, ed. A full modelestimation approach for covariate




effects: Inference based on clinicalimportance and estimation precision.American Association ofPharmaceutical Scientists AnnualMeeting and Exposition; 7–11November 2004; Baltimore, MD: AAPSJournal, 2004.

15. Gibiansky L, Gibiansky E. Target-mediated drug disposition model:relationships with indirect responsemodels and application to populationPK-PD analysis. J PharmacokinetPharmacodyn 2009; 36: 341–51.

16. Tabrizi MA, Tseng CM, Roskos LK.Elimination mechanisms of therapeuticmonoclonal antibodies. Drug DiscovToday 2006; 11: 81–8.

17. Dirks NL, Nolting A, Kovar A,Meibohm B. Populationpharmacokinetics of cetuximab inpatients with squamous cell carcinomaof the head and neck. J Clin Pharmacol2008; 48: 267–78.

18. Frey N, Grange S, Woodworth T.Population pharmacokinetic analysis oftocilizumab in patients with rheumatoidarthritis. J Clin Pharmacol 2010; 50:754–66.

19. Ng CM, Joshi A, Dedrick RL, GarovoyMR, Bauer RJ. Pharmacokinetic-pharmacodynamic-efficacy analysis ofefalizumab in patients with moderate tosevere psoriasis. Pharm Res 2005; 22:1088–100.

20. Xu Z, Seitz K, Fasanmade A, et al.Population pharmacokinetics ofinfliximab in patients with ankylosingspondylitis. J Clin Pharmacol 2008; 48:681–95.

21. Fasanmade AA, Adedokun OJ, Ford J,et al. Population pharmacokinetic

analysis of infliximab in patients withulcerative colitis. Eur J Clin Pharmacol2009; 65: 1211–28.

22. Zhou H, Jang H, Fleischmann RM,et al. Pharmacokinetics and safety ofgolimumab, a fully human anti-TNF-alpha monoclonal antibody, in subjectswith rheumatoid arthritis. J ClinPharmacol 2007; 47: 383–96.

23. Colombel JF, Feagan B, Sandborn W,Assche Gv, Robinson A. Therapeuticdrug monitoring of biologics forinflammatory bowel disease. InflammBowel Dis 2012; 18: 349–58.

24. Nestorov I. Clinical pharmacokineticsof TNF antagonists: how do they differ?Semin Arthritis Rheum 2005; 5(Suppl.1): 12–8.

25. Dirks NL, Meibohm B. Populationpharmacokinetics of therapeuticmonoclonal antibodies. ClinPharmacokinet 2010; 49: 633–59.

26. Fasanmade AA, Adedokun OJ, OlsonA, Strauss R, Davis HM. Serumalbumin concentration: a predictivefactor of infliximab pharmacokineticsand clinical response in patients withulcerative colitis. Int J Clin PharmacolTher 2010; 48: 297–308.

27. Brandse JF, van der Kleij D, WolbinkGJ, et al. The pharmacokinetics ofinfliximab induction therapy in patientswith moderate to severe ulcerativecolitis. Gastroenterology 2014; 146(5Suppl. 1): S134.

28. Van Moerkercke W, Ackaert C,Compernolle G, et al. High infliximablevels are associated with mucosalhealing in Crohn’s disease.Gastroenterology 2010; 138(Suppl. 1):S60.

29. Roblin X, Marotte H, Rinaudo M, et al.Association between pharmacokineticsof adalimumab and mucosal healing inpatients with inflammatory boweldiseases. Clin Gastroenterol Hepatol2014; 12: 80–4.

30. Colombel JF, Sandborn WJ, Allez M,et al. Association between plasmaconcentrations of certolizumab pegoland endoscopic outcomes of patientswith Crohn’s disease. Clin GastroenterolHepatol 2014; 12: 423–31.

31. Wade JR, Parker G, Kosutic G, et al.Population pharmacokinetic analysis ofcertolizumab pegol in patients withCrohn’s disease. J Clin Pharmacol 2015[Epub ahead of print].

32. Velagapudi RB, Noertersheuser P, AwniW, et al. , eds. Effect of methotrexate(MTX) coadministration on thepharmacokinetics (PK) of adalimumab(HUMIRA (TM), Abbott) following asingle intravenous (iv) injection.American College of Rheumatology67th Annual Meeting; 23–28 October2003; Orlando, FL: Arthritis Rheum,2003.

33. Chow A, Earp J, Gupta M, et al. Utilityof population pharmacokineticmodeling in the assessment oftherapeutic protein-drug interactions. JClin Pharmacol 2014; 54: 593–601.

34. Earp J, Fang L, Ma L, et al. Assessinglabeling claims for drug interactionsusing a population PK approach:Vedolizumab. J PharmacokinetPharmacodyn 2014; 41: S39.



M. Rosario et al.

population pharmacokinetics‐pharmacodynamics of ... · population...

Documents